@tangle-network/agent-eval 1.0.1 → 1.0.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/dist/cli.js CHANGED
@@ -9,7 +9,7 @@ import { t as createChatClient } from "./chat-client-D0me8iwq.js";
9
9
  import { t as runAnalystBenchmarkCommand } from "./benchmark-command-BeDZB3BQ.js";
10
10
  import { i as buildOpenApi, l as handleVersion, o as runRpcBatch, r as startServerAsync, s as runRpcOnce } from "./server-CPKyf3bz.js";
11
11
  import { b as editCreditText, f as landscape, g as surfaceTextEdits, l as metaSearch, m as formatLandscape, n as taskMatrix, o as skillManifold, r as formatSkillManifold, s as renderMetaSearchText, t as tree, v as front, y as editCredit } from "./tree-C2N31CoI.js";
12
- import { c as discoveryRunRolloutLines, i as analyzeSupervisorRun, n as renderSupervisorRunHeadline, r as renderSupervisorRunMarkdown } from "./render-aB58Be3N.js";
12
+ import { c as discoveryRunRolloutLines, i as analyzeSupervisorRun, n as renderSupervisorRunHeadline, r as renderSupervisorRunMarkdown } from "./render-CFrJ7M0o.js";
13
13
  import { createHash } from "node:crypto";
14
14
  import { existsSync, readFileSync, writeFileSync } from "node:fs";
15
15
  import { dirname, join, resolve } from "node:path";
@@ -2,8 +2,8 @@ import { o as FailureClass } from "./schema-CkIvnmLU.js";
2
2
  import { i as AnalystFinding, m as EvidenceRef, u as AnalystRunResult } from "./types-DzCO-S1m.js";
3
3
  import { s as TraceStore } from "./store-CMs5wGXP.js";
4
4
  import { i as DatasetSplit, r as DatasetScenario } from "./dataset-DThC1S6G.js";
5
- import "./exact-types-CIViGj98.js";
6
5
  import { o as TraceEmitter } from "./emitter-mAXC5PSN.js";
6
+ import "./exact-types-CIViGj98.js";
7
7
  type ControlSeverity = 'info' | 'warning' | 'error' | 'critical';
8
8
  type ControlActionFailureMode = 'continue' | 'stop';
9
9
  interface ControlEvalResult {
@@ -383,4 +383,4 @@ declare function controlRunToFeedbackTrajectory<TState, TAction, TActionResult>(
383
383
  createdAt?: string;
384
384
  }): FeedbackTrajectory;
385
385
  export { AnalystFindingDigest as A, ControlRuntimeConfig as B, createFeedbackTrajectory as C, renderPreferenceMemoryMarkdown as D, feedbackTrajectoryToOptimizerRow as E, ControlBudget as F, ControlSeverity as H, ControlContext as I, ControlDecision as L, AnalystRunDigest as M, analystRunDigest as N, summarizePreferenceMemory as O, ControlActionOutcome as P, ControlEvalResult as R, controlRunToFeedbackTrajectory as S, feedbackTrajectoriesToOptimizerRows as T, ControlStep as U, ControlRuntimeError as V, StopDecision as W, PreferenceMemoryEntry as _, FeedbackLabel as a, analystRunToReviewRequests as b, FeedbackOptimizerRow as c, FeedbackTask as d, FeedbackTrajectory as f, InMemoryFeedbackTrajectoryStore as g, FileSystemFeedbackTrajectoryStore as h, FeedbackAttempt as i, AnalystReviewDecision as j, withAssignedFeedbackSplit as k, FeedbackOutcome as l, FeedbackTrajectoryStore as m, AnalystReviewRequest as n, FeedbackLabelKind as o, FeedbackTrajectoryFilter as p, FeedbackArtifactType as r, FeedbackLabelSource as s, AnalystFeedbackTrajectoryOptions as t, FeedbackSplitPolicy as u, ProposedSideEffect as v, feedbackTrajectoriesToDatasetScenarios as w, assignFeedbackSplit as x, analystRunToFeedbackTrajectory as y, ControlRunResult as z };
386
- //# sourceMappingURL=feedback-trajectory-Ck7K7AhK.d.ts.map
386
+ //# sourceMappingURL=feedback-trajectory-BgomnFOe.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"feedback-trajectory-Ck7K7AhK.d.ts","names":[],"sources":["../src/control-runtime/contract.ts","../src/feedback-trajectory-review.ts","../src/feedback-trajectory.ts"],"mappings":";;;;;;KAIY;KAEA;UAEK;;EAEf;;EAEA;;EAEA;;EAEA,WAAW;;EAEX;;EAEA;;EAEA;;EAEA,WAAW;;UAGI;EACf;EACA;EACA;;UAGe,oBAAoB,QAAQ;;;;;EAK3C;;;;;EAKA;;EAEA;;EAEA,oBAAoB,OAAO;;EAE3B,qBAAqB,QAAQ;;UAGd,eACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,OAAO;EACP,OAAO;EACP,SAAS,YAAY,QAAQ,SAAS,eAAe;EACrD,QAAQ;EACR;EACA;EACA;EACA;EACA,aAAa;EACb,UAAU;;KAGA,gBAAgB;EAEtB;EACA,QAAQ;EACR;;EAGA;EACA;EACA;EACA;;EAEA,eAAe;;UAGJ;EACf;EACA;EACA;EACA;EACA,eAAe;;UAGA,qBAAqB;EACpC;EACA,SAAS;EACT;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe,YACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,UAAU,gBAAgB;EAC1B,aAAa;EACb,YAAY;EACZ,aAAa;EACb,YAAY;EACZ,gBAAgB,qBAAqB;EACrC;EACA;;UAGe,iBACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA;EACA;EACA;EACA;EACA,OAAO,YAAY,QAAQ,SAAS,eAAe;EACnD,YAAY;EACZ,YAAY;EACZ;EACA;;EAEA;EACA,eAAe;EACf,eAAe;EACf;;UAGe,qBACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,SAAS,QAAQ;EACjB,SAAS;;EAET,gBAAgB;;;;;EAKhB,oBAAoB;IAClB,QAAQ;IACR,QAAQ;IACR,OAAO;IACP,OAAO;IACP,SAAS,YAAY,QAAQ,SAAS,eAAe;;;EAIvD,UAAU;IACR,SAAS,YAAY,QAAQ,SAAS,eAAe;IACrD,aAAa;QACT,QAAQ,UAAU;;EAGxB,WAAW;IACT;IACA,OAAO;IACP,SAAS,YAAY,QAAQ,SAAS,eAAe;IACrD,aAAa;QACT,QAAQ,WAAW;;EAGzB,SACE,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,gBAAgB,YAAY,gBAAgB;;EAGzD,MACE,QAAQ,SACR,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,iBAAiB;;EAG9B,cACE,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,gBAAgB;;EAG7B,UAAU,MAAM,YAAY,QAAQ,SAAS,eAAe,WAAW;;EAGvE,eAAe,oBAAoB,QAAQ;;EAG3C,QAAQ;EACR;EACA;EACA;;KCtMG,sBAAsB,QAAQ;KACvB;KACA;UAEK;EACf;EACA;EACA,WAAW;;UAGH;EACR,WAAW;EACX,QAAQ;EACR;EACA;EACA;EACA;;KAGU,yBACP;EACC;EACA,eAAe;EACf;EACA;MAED;EACC;EACA,cAAc;EACd;EACA;;;iBAgCU,iBAAiB,KAAK,mBAAmB;KCrD7C;KAWA;KAEA;KAYA;UAEK;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA,WAAW;;UAGI;EACf;EACA,QAAQ;EACR,MAAM;EACN;EACA;EACA,WAAW;EACX;EACA,WAAW;;UAGI;EACf;EACA;EACA,cAAc;EACd;EACA;EACA,iBAAiB;EACjB,QAAQ;EACR,WAAW;EACX;EACA,WAAW;;UAGI;EACf;EACA;EACA,UAAU;EACV;EACA;EACA;EACA,WAAW;;UAGI;EACf;EACA;EACA;EACA,MAAM;EACN,UAAU;EACV,QAAQ;EACR,UAAU;EACV,QAAQ;EACR,OAAO;EACP;EACA;EACA,WAAW;;UAGI;EACf,KAAK,YAAY,qBAAqB;EACtC,IAAI,aAAa,QAAQ;EACzB,KAAK,SAAS,2BAA2B,QAAQ;EACjD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAC7D,YAAY,YAAY,OAAO,eAAe,qBAAqB,QAAQ;;UAG5D;EACf;EACA;EACA,QAAQ;EACR;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,YAAY;EACZ;EACA,WAAW;;UAoBI;EACf,MAAM;EACN,kBAAkB;EAClB,0BAA0B;EAC1B,2BAA2B;EAC3B,UAAU;EACV;EACA;EACA;EACA,QAAQ;EACR,OAAO;EACP;EACA;IACE;IACA;IACA;;EAEF,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,UAAU;EACV;EACA;EACA,UAAU;EACV;EACA;;cAUW,2CAA2C;mBACrC;EAEX,KAAK,YAAY,qBAAqB;EAItC,IAAI,aAAa,QAAQ;EAKzB,KAAK,SAAQ,2BAAgC,QAAQ;EAMrD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAa7D,YACJ,YACA,OAAO,eACP,qBACC,QAAQ;;cAsBA,6CAA6C;mBACvC;mBACA;UACT;EAEI,YAAA;IAAW;;EAIjB,KAAK,YAAY,qBAAqB;EAMtC,IAAI,aAAa,QAAQ;EAKzB,KAAK,SAAQ,2BAAgC,QAAQ;EAKrD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAO7D,YACJ,YACA,OAAO,eACP,qBACC,QAAQ;UAOG;UAWA;;iBA8BA,yBAAyB;EACvC;EACA;EACA;EACA,MAAM;EACN,WAAW;EACX,SAAS;EACT,UAAU;EACV,QAAQ;EACR,OAAO;EACP;EACA,WAAW;IACT;;iBAqBY,+BACd,KAAK,kBACL,SAAS,mCACR;;iBAsFa,2BACd,KAAK,kBACL;EAAW;IACV;iBA4Da,oBACd,YAAY,KAAK,iEACjB,SAAQ,sBACP;iBAca,0BACd,YAAY,oBACZ,SAAS,sBACR;iBAuBa,uCACd,cAAc,uBACb;iBAIa,iCACd,YAAY,qBACX;iBA6Ca,oCACd,cAAc,uBACb;iBAyCa,0BACd,cAAc,sBACd;EAAW;IACV;iBA2Ba,+BAA+B,SAAS;iBAWxC,+BAA+B,QAAQ,SAAS,eAC9D,KAAK,iBAAiB,QAAQ,SAAS,gBACvC;EACE;EACA;EACA,eAAe;EACf,oBAAoB,MAAM,YAAY,QAAQ,SAAS;EACvD,0BACE,MAAM,YAAY,QAAQ,SAAS,mBAChC;EACL;IAED"}
1
+ {"version":3,"file":"feedback-trajectory-BgomnFOe.d.ts","names":[],"sources":["../src/control-runtime/contract.ts","../src/feedback-trajectory-review.ts","../src/feedback-trajectory.ts"],"mappings":";;;;;;KAIY;KAEA;UAEK;;EAEf;;EAEA;;EAEA;;EAEA,WAAW;;EAEX;;EAEA;;EAEA;;EAEA,WAAW;;UAGI;EACf;EACA;EACA;;UAGe,oBAAoB,QAAQ;;;;;EAK3C;;;;;EAKA;;EAEA;;EAEA,oBAAoB,OAAO;;EAE3B,qBAAqB,QAAQ;;UAGd,eACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,OAAO;EACP,OAAO;EACP,SAAS,YAAY,QAAQ,SAAS,eAAe;EACrD,QAAQ;EACR;EACA;EACA;EACA;EACA,aAAa;EACb,UAAU;;KAGA,gBAAgB;EAEtB;EACA,QAAQ;EACR;;EAGA;EACA;EACA;EACA;;EAEA,eAAe;;UAGJ;EACf;EACA;EACA;EACA;EACA,eAAe;;UAGA,qBAAqB;EACpC;EACA,SAAS;EACT;EACA;EACA;;UAGe;EACf;EACA;EACA;;UAGe,YACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,UAAU,gBAAgB;EAC1B,aAAa;EACb,YAAY;EACZ,aAAa;EACb,YAAY;EACZ,gBAAgB,qBAAqB;EACrC;EACA;;UAGe,iBACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA;EACA;EACA;EACA;EACA,OAAO,YAAY,QAAQ,SAAS,eAAe;EACnD,YAAY;EACZ,YAAY;EACZ;EACA;;EAEA;EACA,eAAe;EACf,eAAe;EACf;;UAGe,qBACf,QACA,SACA,eACA,cAAc,oBAAoB;EAElC;EACA,SAAS,QAAQ;EACjB,SAAS;;EAET,gBAAgB;;;;;EAKhB,oBAAoB;IAClB,QAAQ;IACR,QAAQ;IACR,OAAO;IACP,OAAO;IACP,SAAS,YAAY,QAAQ,SAAS,eAAe;;;EAIvD,UAAU;IACR,SAAS,YAAY,QAAQ,SAAS,eAAe;IACrD,aAAa;QACT,QAAQ,UAAU;;EAGxB,WAAW;IACT;IACA,OAAO;IACP,SAAS,YAAY,QAAQ,SAAS,eAAe;IACrD,aAAa;QACT,QAAQ,WAAW;;EAGzB,SACE,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,gBAAgB,YAAY,gBAAgB;;EAGzD,MACE,QAAQ,SACR,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,iBAAiB;;EAG9B,cACE,KAAK,eAAe,QAAQ,SAAS,eAAe,WACjD,QAAQ,gBAAgB;;EAG7B,UAAU,MAAM,YAAY,QAAQ,SAAS,eAAe,WAAW;;EAGvE,eAAe,oBAAoB,QAAQ;;EAG3C,QAAQ;EACR;EACA;EACA;;KCtMG,sBAAsB,QAAQ;KACvB;KACA;UAEK;EACf;EACA;EACA,WAAW;;UAGH;EACR,WAAW;EACX,QAAQ;EACR;EACA;EACA;EACA;;KAGU,yBACP;EACC;EACA,eAAe;EACf;EACA;MAED;EACC;EACA,cAAc;EACd;EACA;;;iBAgCU,iBAAiB,KAAK,mBAAmB;KCrD7C;KAWA;KAEA;KAYA;UAEK;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA,WAAW;;UAGI;EACf;EACA,QAAQ;EACR,MAAM;EACN;EACA;EACA,WAAW;EACX;EACA,WAAW;;UAGI;EACf;EACA;EACA,cAAc;EACd;EACA;EACA,iBAAiB;EACjB,QAAQ;EACR,WAAW;EACX;EACA,WAAW;;UAGI;EACf;EACA;EACA,UAAU;EACV;EACA;EACA;EACA,WAAW;;UAGI;EACf;EACA;EACA;EACA,MAAM;EACN,UAAU;EACV,QAAQ;EACR,UAAU;EACV,QAAQ;EACR,OAAO;EACP;EACA;EACA,WAAW;;UAGI;EACf,KAAK,YAAY,qBAAqB;EACtC,IAAI,aAAa,QAAQ;EACzB,KAAK,SAAS,2BAA2B,QAAQ;EACjD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAC7D,YAAY,YAAY,OAAO,eAAe,qBAAqB,QAAQ;;UAG5D;EACf;EACA;EACA,QAAQ;EACR;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,YAAY;EACZ;EACA,WAAW;;UAoBI;EACf,MAAM;EACN,kBAAkB;EAClB,0BAA0B;EAC1B,2BAA2B;EAC3B,UAAU;EACV;EACA;EACA;EACA,QAAQ;EACR,OAAO;EACP;EACA;IACE;IACA;IACA;;EAEF,WAAW;;UAGI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,UAAU;EACV;EACA;EACA,UAAU;EACV;EACA;;cAUW,2CAA2C;mBACrC;EAEX,KAAK,YAAY,qBAAqB;EAItC,IAAI,aAAa,QAAQ;EAKzB,KAAK,SAAQ,2BAAgC,QAAQ;EAMrD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAa7D,YACJ,YACA,OAAO,eACP,qBACC,QAAQ;;cAsBA,6CAA6C;mBACvC;mBACA;UACT;EAEI,YAAA;IAAW;;EAIjB,KAAK,YAAY,qBAAqB;EAMtC,IAAI,aAAa,QAAQ;EAKzB,KAAK,SAAQ,2BAAgC,QAAQ;EAKrD,cAAc,YAAY,SAAS,kBAAkB,QAAQ;EAO7D,YACJ,YACA,OAAO,eACP,qBACC,QAAQ;UAOG;UAWA;;iBA8BA,yBAAyB;EACvC;EACA;EACA;EACA,MAAM;EACN,WAAW;EACX,SAAS;EACT,UAAU;EACV,QAAQ;EACR,OAAO;EACP;EACA,WAAW;IACT;;iBAqBY,+BACd,KAAK,kBACL,SAAS,mCACR;;iBAsFa,2BACd,KAAK,kBACL;EAAW;IACV;iBA4Da,oBACd,YAAY,KAAK,iEACjB,SAAQ,sBACP;iBAca,0BACd,YAAY,oBACZ,SAAS,sBACR;iBAuBa,uCACd,cAAc,uBACb;iBAIa,iCACd,YAAY,qBACX;iBA6Ca,oCACd,cAAc,uBACb;iBAyCa,0BACd,cAAc,sBACd;EAAW;IACV;iBA2Ba,+BAA+B,SAAS;iBAWxC,+BAA+B,QAAQ,SAAS,eAC9D,KAAK,iBAAiB,QAAQ,SAAS,gBACvC;EACE;EACA;EACA,eAAe;EACf,oBAAoB,MAAM,YAAY,QAAQ,SAAS;EACvD,0BACE,MAAM,YAAY,QAAQ,SAAS,mBAChC;EACL;IAED"}
package/dist/index.d.ts CHANGED
@@ -24,6 +24,8 @@ import { a as MultiLayerVerifier, c as VerifyOptions, i as LayerStatus, l as gra
24
24
  import { C as HeldOutGateRejectionCode, S as HeldOutGateDecision, _ as gainHistogram, a as ResearchReportOptions, b as HeldOutGate, d as GainDistributionOptions, g as SummaryTableRow, h as SummaryTableOptions, l as GainDistributionBin, m as SummaryTable, p as ParetoPoint, t as ResearchReport, u as GainDistributionFigureSpec, v as summaryTable, w as SplitCoverage, x as HeldOutGateConfig, y as GateEvidence } from "./summary-report-0La5vejY.js";
25
25
  import { a as FailureClassification, c as FailureRule, d as failureBlame, i as FailureBlame, l as classifyFailure, o as FailureContext, r as failureClusterView, s as FailureReasonRule, u as classifyFailureReason } from "./failure-cluster-DOoa7CBg.js";
26
26
  import { d as InsightReport, n as analyzeRuns, t as AnalyzeRunsOptions } from "./analyze-runs-BYyMShsj.js";
27
+ import { a as SpanHandle, n as CaptureWrite, o as TraceEmitter, s as TraceEmitterOptions, t as CaptureErrorHandler } from "./emitter-mAXC5PSN.js";
28
+ import { s as SandboxDriver } from "./sandbox-harness--VUvMji7.js";
27
29
  import { a as ProducedProposal, c as SatisfiedBy, d as createLlmCorrectnessChecker, f as verifyCompletion, i as LlmCorrectnessCheckerOpts, l as TaskGold, n as CompletionVerdict, o as ProducedState, r as CorrectnessChecker, s as RequirementCheck, t as CompletionRequirement, u as completionVerdict } from "./completion-verifier-7KEfroqx.js";
28
30
  import { _ as SelfImproveMethodResult, a as DefinedAgentEval, c as SelfImproveMethodOptions, d as SelfImproveProposerOptions, f as SelfImproveProposerResult, g as SelfImproveMethodProvenance, h as selfImprove, i as DefineAgentEvalOptions, l as SelfImproveOptions, n as AgentEvalEvaluateOptions, o as defineAgentEval, p as SelfImproveResult, r as AgentEvalImproveOptions, s as SelfImproveBudget, t as AgentEvalAgent, u as SelfImproveProgressEvent } from "./define-agent-eval-B9zPWFSK.js";
29
31
  import { a as createTraceAnalyst, i as TraceAnalystDefinition, n as buildDefaultAnalystRegistry, r as CreateTraceAnalystOptions, t as DefaultAnalystRegistryOptions } from "./default-registry-DKlKVXMx.js";
@@ -33,12 +35,10 @@ import { r as TraceAnalysisEngineResult, s as RawAnalystFinding, t as TraceAnaly
33
35
  import { a as BudgetPolicy, c as RegistryRunOpts, i as AnalystRegistryOptions, s as ExactRegistryRunOpts, t as AnalystRegistry } from "./registry-CMtGAxOw.js";
34
36
  import { C as SEMANTIC_CONCEPT_JUDGE_VERSION, D as runSemanticConceptJudge, E as SemanticConceptJudgeResult, S as ConceptSpec, T as SemanticConceptJudgeOptions, _ as DspyRlmTraceEngineOptions, a as FAILURE_MODE_KIND_SPEC, b as createChatTraceEngine, d as FindingSubject, f as FindingSubjectKind, t as DEFAULT_TRACE_ANALYST_KINDS, v as createDspyRlmTraceEngine, w as SemanticConceptJudgeInput, x as ConceptFinding, y as ChatTraceEngineOptions } from "./index-Bn-P_XAl.js";
35
37
  import { i as BenchmarkEvaluation } from "./types-yofaDmQa.js";
36
- import { a as SpanHandle, n as CaptureWrite, o as TraceEmitter, s as TraceEmitterOptions, t as CaptureErrorHandler } from "./emitter-mAXC5PSN.js";
37
- import { n as SandboxDriver } from "./sandbox-harness-B3Kx1fOA.js";
38
38
  import { _ as redactText, d as ShareSafetyVerdict, f as assessShareSafety, g as redactForShare, h as redact, i as RedactOptions, o as RedactionProfile, r as REDACTION_VERSION, s as RedactionReport, u as ShareSafetyStatus, v as shareAllowed } from "./redact-jrT0APTU.js";
39
39
  import { a as PairedEvalueStep, c as SequentialDecision, i as PairedEvalueSequence, l as evaluateInterimReleaseConfidence, n as InterimReleaseConfidenceInput, r as PairedEvalueOptions, t as InterimReleaseConfidence, u as pairedEvalueSequence } from "./sequential-Bb_pdTqG.js";
40
40
  import { S as welchsTTest, _ as computeToolUseMetrics, b as WelchTestStatus, d as Trajectory, f as TrajectoryStep, g as ToolUseOptions, h as ToolUseMetrics, i as toolWasteView, m as ToolStats, p as buildTrajectory, s as judgeAgreementView, u as budgetBreachView, x as iqr, y as WelchTestResult } from "./tool-waste-BFG1vwTV.js";
41
- import { A as AnalystFindingDigest, B as ControlRuntimeConfig, C as createFeedbackTrajectory, D as renderPreferenceMemoryMarkdown, E as feedbackTrajectoryToOptimizerRow, F as ControlBudget, H as ControlSeverity, I as ControlContext, L as ControlDecision, M as AnalystRunDigest, N as analystRunDigest, O as summarizePreferenceMemory, P as ControlActionOutcome, R as ControlEvalResult, S as controlRunToFeedbackTrajectory, T as feedbackTrajectoriesToOptimizerRows, U as ControlStep, V as ControlRuntimeError, W as StopDecision, _ as PreferenceMemoryEntry, a as FeedbackLabel, b as analystRunToReviewRequests, c as FeedbackOptimizerRow, d as FeedbackTask, f as FeedbackTrajectory, g as InMemoryFeedbackTrajectoryStore, h as FileSystemFeedbackTrajectoryStore, i as FeedbackAttempt, j as AnalystReviewDecision, k as withAssignedFeedbackSplit, l as FeedbackOutcome, m as FeedbackTrajectoryStore, n as AnalystReviewRequest, o as FeedbackLabelKind, p as FeedbackTrajectoryFilter, r as FeedbackArtifactType, s as FeedbackLabelSource, t as AnalystFeedbackTrajectoryOptions, u as FeedbackSplitPolicy, v as ProposedSideEffect, w as feedbackTrajectoriesToDatasetScenarios, x as assignFeedbackSplit, y as analystRunToFeedbackTrajectory, z as ControlRunResult } from "./feedback-trajectory-Ck7K7AhK.js";
41
+ import { A as AnalystFindingDigest, B as ControlRuntimeConfig, C as createFeedbackTrajectory, D as renderPreferenceMemoryMarkdown, E as feedbackTrajectoryToOptimizerRow, F as ControlBudget, H as ControlSeverity, I as ControlContext, L as ControlDecision, M as AnalystRunDigest, N as analystRunDigest, O as summarizePreferenceMemory, P as ControlActionOutcome, R as ControlEvalResult, S as controlRunToFeedbackTrajectory, T as feedbackTrajectoriesToOptimizerRows, U as ControlStep, V as ControlRuntimeError, W as StopDecision, _ as PreferenceMemoryEntry, a as FeedbackLabel, b as analystRunToReviewRequests, c as FeedbackOptimizerRow, d as FeedbackTask, f as FeedbackTrajectory, g as InMemoryFeedbackTrajectoryStore, h as FileSystemFeedbackTrajectoryStore, i as FeedbackAttempt, j as AnalystReviewDecision, k as withAssignedFeedbackSplit, l as FeedbackOutcome, m as FeedbackTrajectoryStore, n as AnalystReviewRequest, o as FeedbackLabelKind, p as FeedbackTrajectoryFilter, r as FeedbackArtifactType, s as FeedbackLabelSource, t as AnalystFeedbackTrajectoryOptions, u as FeedbackSplitPolicy, v as ProposedSideEffect, w as feedbackTrajectoriesToDatasetScenarios, x as assignFeedbackSplit, y as analystRunToFeedbackTrajectory, z as ControlRunResult } from "./feedback-trajectory-BgomnFOe.js";
42
42
  import { A as TraceInsightReadiness, F as describeTraceInsightScope, G as AnalyzeTracesResult, H as tokenizeDomainWords, I as domainEvidencePattern, K as analyzeTraces, L as inferDomainKeywords, N as buildTraceInsightContext, P as buildTraceInsightPrompt, V as scoreTraceInsightReadiness, a as captureFetchToRawSink, b as OtlpSpan, c as ExtractedUsage, d as extractUsageFromSse, f as OtlpFileTraceStore, g as spanRecordsToTraceAnalysisStore, h as otlpTextToTraceAnalysisStore, j as TraceInsightSuite, m as SpanRecordsToTraceAnalysisStoreOptions, n as toolSpansToTraceAnalysisStore, o as OtlpFlatLine, u as extractUsage, v as OtlpExport, x as exportRunAsOtlp } from "./store-tool-spans-lWf6INTH.js";
43
43
  import { a as RunIntegrityReport, o as assertRunCaptured, t as RunIntegrityError } from "./integrity-BAdVzT8u.js";
44
44
  import { a as SteeringChange, c as CampaignRunContext, d as CampaignVariant, f as EvalCampaignOptions, h as runEvalCampaign, i as Researcher, l as CampaignRunOutcome, m as FailedRun, n as ExperimentResult, o as CampaignFactoryParams, p as EvalCampaignResult, r as FailureMode, s as CampaignIntegrityPolicy, t as ExperimentPlan, u as CampaignScenario } from "./researcher-CLnNJ7Or.js";
package/dist/openapi.json CHANGED
@@ -2,7 +2,7 @@
2
2
  "openapi": "3.1.0",
3
3
  "info": {
4
4
  "title": "@tangle-network/agent-eval — wire protocol",
5
- "version": "1.0.1",
5
+ "version": "1.0.3",
6
6
  "description": "HTTP and stdio RPC interface to agent-eval. The TypeScript runtime is the source of truth; this spec is the contract that cross-language clients (Python, Rust, Go) generate from.\n\nWire-protocol version: 1.1.0. Bumps on breaking changes to request/response schemas.",
7
7
  "contact": {
8
8
  "name": "Tangle Network",
@@ -682,16 +682,17 @@ function analyzeSupervisorRunSources(src, now = Date.now) {
682
682
  * analyst reads communication and steer edges from it.
683
683
  */
684
684
  const COORDINATION_FILE = "coordination-log.jsonl";
685
+ const readFileText = (path) => readFile(path, "utf8");
685
686
  /** File text, or null when the file does not exist. Any other read error throws. */
686
- async function readMaybe(path) {
687
- return readFile(path, "utf8").catch((error) => {
687
+ async function readMaybe(path, read = readFileText) {
688
+ return read(path).catch((error) => {
688
689
  if (typeof error === "object" && error !== null && "code" in error && error.code === "ENOENT") return null;
689
690
  throw error;
690
691
  });
691
692
  }
692
693
  /** The run's coordination log, or null when Runtime wrote none. */
693
- function readCoordinationLog(runDir) {
694
- return readMaybe(join(runDir, COORDINATION_FILE));
694
+ function readCoordinationLog(runDir, read = readFileText) {
695
+ return readMaybe(join(runDir, COORDINATION_FILE), read);
695
696
  }
696
697
  // src/supervisor-run/runtime-source-limits.ts
697
698
  function spendRecord(value) {
@@ -1015,10 +1016,10 @@ const HARNESS_SESSION_FILE = {
1015
1016
  opencode: /(^|\/)(\.local\/share\/opencode\/(opencode\.db|storage\/session\/.+\.json|export\/.+\.json)|\.opencode\/sessions\/.+)$/u
1016
1017
  };
1017
1018
  /** Session files the receipt's transcript blob lists, or undefined when it cannot say. */
1018
- async function receiptSessionFiles(file) {
1019
+ async function receiptSessionFiles(file, read) {
1019
1020
  let descriptor;
1020
1021
  try {
1021
- descriptor = JSON.parse(await readFile(file, "utf8"));
1022
+ descriptor = JSON.parse(await read(file));
1022
1023
  } catch {
1023
1024
  return;
1024
1025
  }
@@ -1055,7 +1056,7 @@ function nativeTraceSummary(workers, root) {
1055
1056
  const gaps = [count("partial") > 0 ? `${count("partial")} only a partial copy` : null, count("no-session-file") > 0 ? `${count("no-session-file")} a receipt with no session file` : null].filter((gap) => gap !== null);
1056
1057
  return `per node — ${count("complete")} of ${workers.length} workers carry a complete native session receipt (workers[].nativeSession)${gaps.length > 0 ? `; ${gaps.join("; ")}` : ""}; root: ${rootText} (result.json rootHarnessTranscript)`;
1057
1058
  }
1058
- async function workspaceCapturesFromOutput(runDir, ref) {
1059
+ async function workspaceCapturesFromOutput(runDir, ref, read) {
1059
1060
  const outRef = typeof ref === "string" ? ref : null;
1060
1061
  const blobPath = blobFile(runDir, ref);
1061
1062
  const missing = (projectionGap) => ({
@@ -1070,7 +1071,7 @@ async function workspaceCapturesFromOutput(runDir, ref) {
1070
1071
  });
1071
1072
  if (blobPath === null) return [missing("Runtime output has no valid content address")];
1072
1073
  if (!await isFile(blobPath)) return [missing("Runtime output blob is missing or not a file")];
1073
- const raw = await readMaybe(blobPath);
1074
+ const raw = await readMaybe(blobPath, read);
1074
1075
  if (raw === null) return [missing("Runtime output blob is missing")];
1075
1076
  let output;
1076
1077
  try {
@@ -1146,7 +1147,7 @@ async function workspaceCapturesFromOutput(runDir, ref) {
1146
1147
  * Blob files are checked on disk: a receipt that names a blob the directory no longer holds is
1147
1148
  * not a retained transcript.
1148
1149
  */
1149
- async function nodeTranscripts(runDir, events, nodeIds, root) {
1150
+ async function nodeTranscripts(runDir, events, nodeIds, read, root) {
1150
1151
  const facts = /* @__PURE__ */ new Map();
1151
1152
  const entry = (id) => {
1152
1153
  let found = facts.get(id);
@@ -1193,14 +1194,14 @@ async function nodeTranscripts(runDir, events, nodeIds, root) {
1193
1194
  retained.push(file);
1194
1195
  if (turn) retainedTurns += 1;
1195
1196
  }
1196
- workspaceCaptures.push(...await workspaceCapturesFromOutput(runDir, ref));
1197
+ workspaceCaptures.push(...await workspaceCapturesFromOutput(runDir, ref, read));
1197
1198
  }
1198
1199
  let nativeSession = null;
1199
1200
  if (fact.receipt !== null) {
1200
1201
  if (fact.receipt.status === "available") {
1201
1202
  const file = blobFile(runDir, fact.receipt.transcriptRef);
1202
1203
  if (file !== null && await isFile(file)) {
1203
- const sessionFiles = await receiptSessionFiles(file);
1204
+ const sessionFiles = await receiptSessionFiles(file, read);
1204
1205
  const coverageComplete = fact.receipt.coverageComplete;
1205
1206
  nativeSession = {
1206
1207
  status: "available",
@@ -1230,11 +1231,6 @@ async function nodeTranscripts(runDir, events, nodeIds, root) {
1230
1231
  return out;
1231
1232
  }
1232
1233
  /**
1233
- * Sources for a run dir whose spawn journal does not exist. Mirrors the
1234
- * absent shape the reader returns for a missing store: every
1235
- * journal-dependent metric downstream reads `unavailable`, never 0.
1236
- */
1237
- /**
1238
1234
  * The run identity a terminal record names when no journal exists yet: the
1239
1235
  * settle record's `tree.root`, else the failure record's `runId`. Both are
1240
1236
  * Runtime's own `runId`, so the report names the run the record is about
@@ -1290,12 +1286,13 @@ function absentRuntimeSupervisorRun(runDir, resultText, failureText, instanceId)
1290
1286
  * verdict.
1291
1287
  */
1292
1288
  async function readRuntimeSupervisorRun(runDir, opts = {}) {
1289
+ const read = opts.readFile ?? readFileText;
1293
1290
  const journalPath = join(runDir, JOURNAL_FILE);
1294
- const rawJournal = opts.strict === true ? await readFile(journalPath, "utf8") : await readMaybe(journalPath);
1291
+ const rawJournal = opts.strict === true ? await read(journalPath) : await readMaybe(journalPath, read);
1295
1292
  const resultPath = join(runDir, RESULT_FILE);
1296
1293
  const failurePath = join(runDir, FAILURE_FILE);
1297
- const resultText = await readMaybe(resultPath);
1298
- const failureText = await readMaybe(failurePath);
1294
+ const resultText = await readMaybe(resultPath, read);
1295
+ const failureText = await readMaybe(failurePath, read);
1299
1296
  const failure = parseOptionalRecord(failureText, failurePath);
1300
1297
  assertFailureRecord(failure, failurePath);
1301
1298
  const result = parseOptionalRecord(resultText, resultPath);
@@ -1304,7 +1301,7 @@ async function readRuntimeSupervisorRun(runDir, opts = {}) {
1304
1301
  assertResultMatchesJournal(normalized.root, result, resultPath);
1305
1302
  const childSpawns = normalized.events.filter((event) => event.kind === "spawned" && nonEmptyString(event.id) !== null).filter((event) => event.id !== normalized.root);
1306
1303
  const workerIds = new Set(childSpawns.map((event) => nonEmptyString(event.id)).filter((id) => id !== null));
1307
- const transcripts = await nodeTranscripts(runDir, normalized.events, /* @__PURE__ */ new Set([normalized.root, ...workerIds]), {
1304
+ const transcripts = await nodeTranscripts(runDir, normalized.events, /* @__PURE__ */ new Set([normalized.root, ...workerIds]), read, {
1308
1305
  id: normalized.root,
1309
1306
  receipt: recordOrNull(result?.rootHarnessTranscript)
1310
1307
  });
@@ -1346,7 +1343,7 @@ async function readRuntimeSupervisorRun(runDir, opts = {}) {
1346
1343
  judgeSource: null,
1347
1344
  patch: null,
1348
1345
  driverLog: null,
1349
- coordinationLog: await readCoordinationLog(runDir),
1346
+ coordinationLog: await readCoordinationLog(runDir, read),
1350
1347
  harnessWorkerTokens: null,
1351
1348
  harnessMissingReason: "Runtime FileRunContext has no external worker-token join",
1352
1349
  limits: sourceLimits(normalized.root, normalized.events, workerIds),
@@ -1365,10 +1362,13 @@ async function isRuntimeSupervisorRunDir(runDir) {
1365
1362
  return false;
1366
1363
  }
1367
1364
  async function isFile(path) {
1368
- return stat(path).then((entry) => entry.isFile()).catch((error) => {
1369
- if (typeof error === "object" && error !== null && "code" in error && (error.code === "ENOENT" || error.code === "ENOTDIR")) return false;
1365
+ try {
1366
+ return (await stat(path)).isFile();
1367
+ } catch (error) {
1368
+ const code = error !== null && typeof error === "object" && "code" in error ? error.code : null;
1369
+ if (code === "ENOENT" || code === "ENOTDIR") return false;
1370
1370
  throw error;
1371
- });
1371
+ }
1372
1372
  }
1373
1373
  // src/supervisor-run/discovery-rollouts.ts
1374
1374
  /**
@@ -1866,4 +1866,4 @@ function renderSupervisorRollupMarkdown(rollup, title = "Round rollup") {
1866
1866
  }
1867
1867
  export { findSupervisorRunDirs as a, discoveryRunRolloutLines as c, analyzeSupervisorRunSources as d, parsePatch as f, analyzeSupervisorRun as i, isRuntimeSupervisorRunDir as l, renderSupervisorRunHeadline as n, discoveryPlayCheckReward as o, renderSupervisorRunMarkdown as r, discoveryPursuitOf as s, renderSupervisorRollupMarkdown as t, readRuntimeSupervisorRun as u };
1868
1868
 
1869
- //# sourceMappingURL=render-aB58Be3N.js.map
1869
+ //# sourceMappingURL=render-CFrJ7M0o.js.map